Chess will never be solved, here's why

There's a variation on what Elroch just said.
Anytime the computer finds a forced checkmate - it doesn't have to find a faster or slower one. That's that. It 'gets a break' on the faster one.

If it finds a forced draw or forced perpetual check or forced stalemate though - it can't stop there. Its not finished that yet.
Because what if there's a forced win for the player supposedly trying to draw?
It has to check thoroughly for that too - even if it doesn't look likely.

Or what about just 'winning chances'? What if the player doesn't want a draw?
That's making it tougher yet again.
I guess we have to give the computer a break and it doesn't have to look for helpmates and helpdraws and helpwins? That's not 'open and shut' either.
In fact - mistakes (help) is what the real game of chess is all about.
Elroch already indirectly suggesting that too in his posts just now.

Its really getting from 7 to 8 pieces that's key - and 8 to 9 and so on up to and including 32.
But the sun's going to engulf the earth long before that job gets done!

@9714

"a weak solution consists of a strategy"
++ Knowledge-based methods are just as acceptable as brute-force methods.
'It is often beneficial to incorporate knowledge-based methods in game-solving programs'
Games solved: Now and in the future

"against ANY opponent moves" ++ Correction: against any opposition.

"it needs to deal with all dumb sequences of opponent moves"
++ No. You can prune. 1 e4 Nf6 2 Qh5? is a loss for white without needing any game tree.

"heuristic analysis like conventional chess engines with their evalutions was used heavily to guide the choice of likely moves to include in the strategies, but played no role in the final proof"
++ I understand this completely: the final proof is reaching the 7-men endgame table base or a prior 3-fold repetition. Engine evaluations play no role.

"The rest can be proven to be irrelevant by an Alpha-Beta search" ++ Same for Chess.

@9716

"0 error: 104 games
1 error: 0 games
2 errors: 1 game"
++ It is conceivable. It is still consistent with Chess is a draw.
So 1 in 105 games could be non-perfect.

"Note that such double errors may be quite common when the players are quite similar,
because they can involve both players missing a very difficult line."
++ Yes, it is a problem for engine autoplay by 1 entity.
However different ICCF grandmasters with different engines and tunings make it unlikely.

"Likely examples are often spotted in world class GM games analysed by top engines."
++ Humans are much weaker. There is also psychology. I make an incorrect sacrifice.
My opponent looks at it, and his clock is ticking, and he wrongfully declines.

@9717

"and so on up to and including 32"
++ That would be strongly solving chess and it is beyond present technology.
Move on to weakly solving, as Schaeffer did for Checkers.

Quite right. What do you think of "weakly solved" as a descriptor?

I think it's obvious jargon and people who aren't too sure of themselves hide behind jargon. They can always claim that an outsider doesn't understand the terms, like Elroch's "Perfect Information", which keeps outsiders out of discussions and helps make sure everyone who doesn't agree with them is an outsider. I think that if someone can't describe something without using jargon, it's a strong sign that they don't properly understand it.

Secondly, a "weak solution" is just that. Aside from the fact that the definition is a heap of nonsense it conveys the idea that it is what it is: a guess, since at present there's no foolproof way to seperate good lines in chess from bad. It's beyond doubt to me that three GMs, holding a seance at midnight next to a bubbling caudron of truth elexir, aren't going to manage.

But have you any positive opinions in any direction?

I could add that I did show that a chess position isn't perfect information, any more than a code for which we don't have the full key, in that the true meanings both of that code and the chess position can't be fully understood, in regard to which the true meaning of a chess position may be abbreviated as won, drawn or lost. Likewise, the code might say that the attack is at dawn, in the evening or just after lunch and at other times it's most important to get some sleep.

I didn't notice a response to that from Elroch: a sure sign that he's gone away to think about it for three days, after which he will return full of confidence, with an answer which consists of a distraction strategy, if the past is to be any guide.

@9721

"even if chess is a draw we don't know that" ++ Yes, we know.

"we don't have proof" ++ We have.

"until engines have solved all 32 positions we don't have proof"
++ This thinking error keeps coming up. It is not necessary to strongly solve Chess to weakly solve it. It is not necessary to weakly solve Chess to ultra-weakly solve it.
We can know Chess is a draw without knowing how.
We can know Chess is a draw and know how to draw Chess without a 32-men table base.

Schaeffer weakly solved Checkers, not strongly. We know Checkers is a draw and know how.
We do not have a full table base for Checkers.

"it's known that chess is a draw" ++ Yes

"just no proof" ++ The 105 ICCF WC Finals draws are the proof.

"how that works" ++ Proof is proof, by whatever means.

"what your trying to do is assume something" ++ No I look at an observed fact: 105 out of 105 ICCF WC Finals games are draws. This is not consistent with Chess not being a draw. It is also not consistent with the 105 games not being perfect games (except maybe one or even two).

@9722

"weak solution is just that" ++ Please do read Games solved: Now and in the future as you promised to do instead of talking nonsense.

"there's no foolproof way to seperate good lines in chess from bad"
++ There is: any line that ends in either a 7-men endgame table base draw,
or a prior 3-fold repetition is a good line for black.

"three GMs" ++ 17 ICCF (grand)masters with engines 5 days/move average now achieved 105 draws out of 105 games.

@9723

"a chess position isn't perfect information"
++ It is: you know all pieces from both sides. It is unlike Bridge, Stratego, Backgammon (you know the pieces, but not the dice), Dominoes, Poker...

@9727

"statistics isn't solid proof 100% of the time" ++ Last year I had to use statistics,
i.e. a Poisson distribution, as there were some decisive games in the ICCF WC Finals.
This year 105 draws out of 105 games, no statistics needed.

"computers aren't smart enough to do that yet"
++ The ICCF grandmasters with their engines now are.

"they like playing a bunch of petrovs which is a draw"
++ There were only a few Petrov Defences.

"What were the openings played?" ++ Ruy Lopez, Italian, Petrov, Najdorf, French, Queen's Gambit, Catalan, Nimzovich Indian Defense, Queen's Indian Defense, Reti...
The interesting thing is that there is more than one way to draw for black.
So even if there were a nonperfect game, there is still a fallback option to draw.

If you get your information out of that book then why should I want to read it? Oh is it 17 GMs now? It was three a couple of years ago wasn't it?

I accept more than Elroch and Megacheese do that a GM with an engine at five days per move is getting close to being very accurate indeed. Maybe we could say "close to perfect accuracy". It means that with such a method we could look at 17 different games simultaneously at one move in five days so discounting openings, let's say of six moves, but given that computer games will be much longer than normal games, maybe middle games are going to be 70 moves long instead of 30 to 35 as is the case with humans?

So 17 games from opening position to getting close to a tablebase position might be 70 x 5 days which is about a year, so that's 17 games a year and to weakly solve chess, do you think we could do it by looking at 1000 different, representative positions? There has to be that sort of overkill. 100 games may be possible but I wouldn't want to bet my house on it. 1000 games should be enough. Unlike Megachess I don't think looking at every possible continuation of every possible continuation is what we need to do.

So that's what, 60 years to get close to "weakly" solving chess using 17 GMs. OK I'll buy that. Actually I think that one computer literate GM with 17 computers would do the same job.

edited^

@9731

"What about the kings gambit" ++ Nobody is foolish enough to play King's Gambit in an ICCF World Championship Finals. It is in all probability a white loss, and even if white manages to draw, it is months and months of suffering. They try to win as white, not draw. Even then they draw.

++ There is: any line that ends in either a 7-men endgame table base draw,
or a prior 3-fold repetition is a good line for black.>>> I'm supposed to be working.

I'm afraid I missed this!! How do you get to the 7-men tablebase draw without doing the 60 year analysis? That's what we're talking about and it means that there's no way to seperate good lines from bad without going through the 60 year process.

In other words, a wide spectrum of lines to guard against errors.

@9732

"If you get your information out of that book then why should I want to read it?"
++ It is a paper, no book, and it would prevent you from talking nonsense.

"Oh is it 17 GMs now? It was three a couple of years ago wasn't it?" ++ I originally calculated 3 grandmasters with 3 supercomputers can weakly solve Chess in 5 years as Sveshnikov predicted. Now however the 17 ICCF (grand)masters with their engines do the job.

"close to perfect accuracy" ++ They came closer to perfect accuracy in previous years,
but now they have reached perfection: 105 draws out of 105 games.

"given that computer games will be much longer than normal games"
++ No. Shortest 15, longest 73, average 38, standard deviation 11.

"which is about a year" ++ They started 20 November 2022. 31 games are still ongoing.
Each of the 17 ICCF grandmasters uses several engines. If I were to qualify for the ICCF World Championship, I would buy, rent, borrow, or steal whatever computer power I could.

"1000 games should be enough."
++ There will be 136 when it is over. Next year another 136 unless they stop organising it.

"I don't think looking at every possible continuation of every possible continuation is what we need to do." ++ Of course not.

Oh OK, with the method you propose I agree that you can weakly solve chess in 5 years probably to a confidence value of 99.9%, only if the average length of games is 76 moves to a tablebase. I've a feeling that they will be longer, however, but I'm prepared to agree that even 15 years is ball-park and a similar result.

I would consider that degree of confidence to be sufficient but would the others??

@9736

"How do you get to the 7-men tablebase draw" ++ White plays what he thinks is his best move. Black plays what he thinks is his best move. They play move after move average 5 days/move, with engines. Then they arrive at a 7 men endgame table base draw, or a prior 3-fold repetition.

"there's no way to seperate good lines from bad"
++ If it ends in a draw, then it was a good line for black.